AI citation checker
Paste a domain and we will tell you whether an AI assistant cited it — if it is one of the 48 pages we actually measured. There is no scan here: we have no server that can fetch your site, so we will not pretend to look. What we can show you is every run, every search rank, and every page fact behind the number.
Looking for APA or MLA formatting? That is a different tool. In academic writing a “citation checker” validates a reference against a style guide. This page answers one question only: did an AI assistant list your page as a source?
The record
The three answers, and what each one actually means
Cited. At least one of the runs listed this URL among the sources it used. You can see which runs and how many. Being cited once out of six is not the same as being cited every time, and the row shows you which one you got.
Measured, never cited. The page was in the pool — the model could read it — and the answer did not use it. This is the more interesting failure, because nothing was blocked and nothing was missing. 26 of our 48 pages are in this bucket.
Not in the 48. We say so, in those words. It does not mean AI ignores your site; it means you were not in the top five results for any of the five queries we ran, so there was nothing for us to measure. That is the other way to fail — never retrieved at all — and it is the one no page-level fix can reach.
What the 48 records show
22 of the 48 pages were cited in at least one run. The other 26 were in the pool every time and were never used once.
Citation rate per run for pages carrying JSON-LD against pages without it — 59 of 164 runs against 11 of 105. The largest gap we found in fifteen page-level features.
9 of the 22 pages that were cited ranked outside the top three in search. Five of the ten pages that ranked first were never cited at all.
Which of these differences is a signal, and which is noise
A gap between two percentages is not automatically a finding. At this sample size — 48 pages — a rank correlation needs |ρ| ≥ 0.285 before it is distinguishable from noise at the usual 5% level, and exactly one of the fifteen features we scored clears that bar: whether the page carried JSON-LD, at ρ 0.391.
The crawler-access gap looks bigger in the raw percentages and is not significant: ρ 0.162 on n = 43, against a threshold of 0.301 for a sample that small. The same goes for llms.txt at 0.168, under the 0.285 line. Both numbers are real. Neither is strong enough to tell you what to change, and we would rather write that sentence than round a 0.168 up into advice.
One more thing the numbers say, and it is the uncomfortable one: search rank barely predicts citation. 9 of the 22 cited pages ranked outside the top three, and 5 of the 10 pages that ranked first were never cited. If ranking were the whole story, those two numbers would be close to zero.
All 48 records
The bar on each row is that page's citation count across runs — not a score, not a ranking, not a prediction. Six cells means six usable runs; a filled cell is a run where the URL appeared in the source list.
What this page cannot tell you
It cannot see your site. There is no fetch, no crawl, no background job. If you paste a domain we never measured, you get a sentence saying we never measured it — that is the entire feature, and it is deliberate.
It is one model, one day. Everything here comes from a single set of 30 runs on 16 October 2026 against one model behind one endpoint. A different model has a different source list. Treat this as a measurement, not as a property of the web.
A single run is noise. Ask the same question twice and the source list changes: run-to-run agreement across our five queries ranged from 0.472 to 0.700. That is why every count on this page is out of five or six runs rather than one, and why a 1/6 is not proof that a page does not work.
Correlation is not causation. Pages with JSON-LD were cited more often. That is consistent with structured data helping a model decide what a page is about, and it is also consistent with the kind of site that bothers to add JSON-LD being the kind of site that does everything else right. We have not separated those two, and we are not going to pretend we have.
Questions about the checker
Does this scan my website?
No. There is no server behind this page — it is a script in your browser reading a fixed set of 48 records we measured on 16 October 2026. If your domain is not in those records, the honest answer is that we do not know, and the page says exactly that instead of showing a progress bar.
What counts as a citation here?
A run counts as citing a page when the assistant listed that page's URL among the sources it used for its answer. Not a mention of the brand in the prose — the URL in the source list. That is stricter than most visibility tools, which count any appearance of a name.
My domain is not in the 48. What does that mean?
It means you did not appear in the top five results for any of the five queries we ran, so you were never in the pool we could measure. That is the first of the two ways to fail: not being retrieved at all. The other is being retrieved and then not used — 26 of our 48 pages were in the pool and were never cited once.
Is this the same as an APA or MLA citation checker?
No, and the two get confused constantly. An academic citation checker validates a reference against a style guide. This page answers a different question: did an AI assistant use my page as a source? We put the difference at the top of the page, because someone searching for the other tool should not have to work it out from the results.
Why only 48 pages?
Because 48 is what one person can measure honestly in a week with a published method. Five queries, the top five organic results for each, one model at temperature 0.7 with a unique nonce per run, six runs per query, 28 of 30 runs usable. Every number on this page recomputes from those runs — the method page carries the script.
Can I pay to appear in these results?
No. The records are a measurement, not a listing, and nothing on this page changes them. We do not sell placement anywhere on the site, and the one place a paid position exists is labelled as paid and marked sponsored.
If you were not in the 48
A lookup that ends in “not found” is not useful on its own. The two things below run entirely in your browser on a page you paste, and they do not depend on our sample at all — one tells you what a machine can and cannot lift off your page, the other writes the files that fix the mechanical half of it.